Repository navigation
Description
1. Executive Summary
Status: HEALTHY
Period: 2026-06-25T16:02:30Z – 2026-06-26T08:35:59Z
Result: 1 of 100 runs failed (1.0%); 23 cancelled (23%), 2 skipped
Key findings:
- 22 of 23 cancellations form a single tight cluster (16:02–16:04 UTC Jun 25) — consistent with a workflow file push triggering GitHub's auto-cancel-superseded-runs behavior, not a recurring outage
- 1 transient failure (#44441 at 19:40:25Z) followed immediately by a cancel and skip, then clean recovery at #44444 — no log payload was captured, preventing root-cause confirmation
- No error logs were returned for the failed run; the "Failed Run Logs" section in the input is empty, so error-level analysis is limited to structural inference
Action required: Confirm whether run #44441's failure was transient (retry succeeded) or indicates a silent error that requires log retention fixes.
2. Failure Breakdown
| Failure Category | Affected Runs | Example Error Message |
|---|---|---|
| Transient / unknown (no logs captured) | 1 (#44441) | (no log output available) |
| Operational cancellation — workflow redeploy | 22 (#44389–#44410) | (cancelled, not failed — GitHub auto-cancel on push) |
| Conditional skip | 2 (#44436, #44443) | (skipped — likely if: branch not taken) |
3. Error Patterns
Transient failure — Run #44441
No log content was returned for the single failing run. The surrounding context is:
#44440 success @ 19:36:33Z
#44441 failure @ 19:40:25Z ← failure
#44442 cancelled @ 19:40:26Z ← concurrent sibling, cancelled 1 second later
#44443 skipped @ 19:40:27Z ← another sibling, skipped 2 seconds later
#44444 success @ 19:40:44Z ← clean recovery 19 seconds after failure
The sibling cancellation/skip pattern (all within ~20 seconds of each other) is consistent with a check_suite or pull_request event fanning out to multiple workflow instances, where the first one fails partway through and GitHub cancels or skips the overlapping duplicates. The rapid recovery without manual intervention suggests the failure was transient (rate limit, ephemeral network error, or a single-step non-zero exit that did not recur).
Mass cancellation — Runs #44389–#44410
#44389–#44410: all cancelled between 16:02:30Z and 16:04:15Z (102-second window)
#44411 success @ 16:04:19Z ← immediate recovery
This pattern is the canonical GitHub Actions response to a workflow file commit on the default branch or a tag move (e.g. pr-review/next being re-pointed by cut-release.sh). GitHub cancels all in-progress runs of the same workflow on the same ref when a new commit supersedes them. Not a fault.
4. Token Scope Analysis
Scopes declared in workflow source:
| Scope | Level | Granted To |
|---|---|---|
contents: read |
job-level | Caller job → passed to reusable |
pull-requests: write |
job-level | Caller job → passed to reusable |
checks: read |
job-level | Caller job → passed to reusable |
Scopes that appear missing or insufficient:
No auth errors were observed in the run summary, and no gh auth status output was included in the logs. Based on the workflow's described function (PR review enumeration, comment posting, check_suite reads), the declared scopes appear sufficient for normal operation.
| Scope to verify | Reason |
|---|---|
checks: read |
Required for check_suite trigger correlation — present, no observed errors |
pull-requests: write |
Required for review submission — present, no observed errors |
metadata: read |
Implicitly granted to all tokens — not a gap |
Recommendation: No scope gaps are confirmed from available data. If #44441's failure ever recurs, capture gh auth status output in the failure step to rule out token expiry or scope regression after a secrets rotation.
5. Recommendations
-
[HIGH] Fix log capture for failed runs.
The "Failed Run Logs" payload for run #44441 is empty. If the log-fetching script (likelyscripts/engine.shor the health-check caller) only fetches logs from jobs in afailureconclusion but the job timed out or was force-killed before GitHub flushed logs, the root cause will never be visible.- What to change: In the health-check or log-fetch script, add
--attemptiteration (GitHub logs each attempt separately) and a fallback to fetching the last N bytes of the step log viagh api repos/{owner}/{repo}/actions/runs/{run_id}/logs. - Why: One silent failure per day compounds into undiagnosable toil over weeks.
- Expected impact: Future failures will have actionable error messages.
- What to change: In the health-check or log-fetch script, add
-
[MEDIUM] Confirm the cancellation cluster was triggered by a tag move, not an unplanned event.
The 22-run cancellation at 16:02–16:04 UTC is consistent with acut-release.shrun or a directgit push --force-with-leasetopr-review/next. Verify against the tag history:gh api repos/petry-projects/.github-private/git/refs/tags/pr-review/next \ --jq '.object.sha' # then cross-reference commit timestamp with 2026-06-25T16:02Z
- Why: If it was an unplanned push, the ring-0 canary may have silently skipped PRs that were open during the cancellation window.
- Expected impact: Either confirms normal release cadence or surfaces a gap in PR review coverage.
-
[LOW] Add a
concurrencygroup withcancel-in-progress: falseguard for the review job.
The sibling cancellation pattern around #44441–#44443 suggests multiple workflow instances fan out on the same event and the first failure may cancel siblings before they can independently succeed.- What to change: In
.github/workflows/pr-review-trigger.yml, consider aconcurrency:block scoped to${{ github.event.pull_request.number || github.event.check_suite.id }}withcancel-in-progress: falseso a failed instance does not suppress sibling retries. - Why: The current behavior may silently swallow a successful retry if GitHub's cancellation races the sibling's success.
- Expected impact: Eliminates the ambiguous cancel/skip tail after a transient failure.
- What to change: In
6. Health Score
Health: 8/10 — one unexplained transient failure with no retained logs; mass cancellations are attributable to a normal release-channel tag move rather than systemic failure.
Activity
don-petry commentedon Jul 14, 2026
Closing stale daily digest — superseded by the newest snapshot. Digest steps now close prior days automatically (#1204 / #1212).